langchainrb 0.6.10 → 0.6.12

data/lib/langchain/prompt/loading.rb

@@ -11,82 +11,88 @@ module Langchain::Prompt
     "few_shot" => ->(config) { load_few_shot_prompt(config) }
   }
 
-  class << self
-    #
-    # Load prompt from file.
-    #
-    # @param file_path [String, Pathname] The path of the file to read the configuration data from.
-    #
-    # @return [Object] The loaded prompt loaded.
-    #
-    # @raise [ArgumentError] If the file type of the specified file path is not supported.
-    #
-    def load_from_path(file_path:)
-      file_path = file_path.is_a?(String) ? Pathname.new(file_path) : file_path
-
-      case file_path.extname
-      when ".json"
-        config = JSON.parse(File.read(file_path))
-      when ".yaml", ".yml"
-        config = YAML.safe_load(File.read(file_path))
-      else
-        raise ArgumentError, "Got unsupported file type #{file_path.extname}"
-      end
-
-      load_from_config(config)
+  module Loading
+    def self.included(base)
+      base.extend ClassMethods
     end
 
-    #
-    # Loads a prompt template with the given configuration.
-    #
-    # @param config [Hash] A hash containing the configuration for the prompt.
-    #
-    # @return [PromptTemplate] The loaded prompt loaded.
-    #
-    def load_prompt(config)
-      template, input_variables = config.values_at("template", "input_variables")
-      PromptTemplate.new(template: template, input_variables: input_variables)
-    end
+    module ClassMethods
+      #
+      # Load prompt from file.
+      #
+      # @param file_path [String, Pathname] The path of the file to read the configuration data from.
+      #
+      # @return [Object] The loaded prompt loaded.
+      #
+      # @raise [ArgumentError] If the file type of the specified file path is not supported.
+      #
+      def load_from_path(file_path:)
+        file_path = file_path.is_a?(String) ? Pathname.new(file_path) : file_path
 
-    #
-    # Loads a prompt template with the given configuration.
-    #
-    # @param config [Hash] A hash containing the configuration for the prompt.
-    #
-    # @return [FewShotPromptTemplate] The loaded prompt loaded.
-    #
-    def load_few_shot_prompt(config)
-      prefix, suffix, example_prompt, examples, input_variables = config.values_at("prefix", "suffix", "example_prompt", "examples", "input_variables")
-      example_prompt = load_prompt(example_prompt)
-      FewShotPromptTemplate.new(prefix: prefix, suffix: suffix, example_prompt: example_prompt, examples: examples, input_variables: input_variables)
-    end
+        case file_path.extname
+        when ".json"
+          config = JSON.parse(File.read(file_path))
+        when ".yaml", ".yml"
+          config = YAML.safe_load(File.read(file_path))
+        else
+          raise ArgumentError, "Got unsupported file type #{file_path.extname}"
+        end
 
-    private
+        load_from_config(config)
+      end
 
-    #
-    # Loads the prompt from the given configuration hash
-    #
-    # @param config [Hash] the configuration hash to load from
-    #
-    # @return [Object] the loaded prompt
-    #
-    # @raise [ArgumentError] if the prompt type specified in the config is not supported
-    #
-    def load_from_config(config)
-      # If `_type` key is not present in the configuration hash, add it with a default value of `prompt`
-      unless config.key?("_type")
-        Langchain.logger.warn "No `_type` key found, defaulting to `prompt`"
-        config["_type"] = "prompt"
+      #
+      # Loads a prompt template with the given configuration.
+      #
+      # @param config [Hash] A hash containing the configuration for the prompt.
+      #
+      # @return [PromptTemplate] The loaded prompt loaded.
+      #
+      def load_prompt(config)
+        template, input_variables = config.values_at("template", "input_variables")
+        PromptTemplate.new(template: template, input_variables: input_variables)
       end
 
-      # If the prompt type specified in the configuration hash is not supported, raise an exception
-      unless TYPE_TO_LOADER.key?(config["_type"])
-        raise ArgumentError, "Loading #{config["_type"]} prompt not supported"
+      #
+      # Loads a prompt template with the given configuration.
+      #
+      # @param config [Hash] A hash containing the configuration for the prompt.
+      #
+      # @return [FewShotPromptTemplate] The loaded prompt loaded.
+      #
+      def load_few_shot_prompt(config)
+        prefix, suffix, example_prompt, examples, input_variables = config.values_at("prefix", "suffix", "example_prompt", "examples", "input_variables")
+        example_prompt = load_prompt(example_prompt)
+        FewShotPromptTemplate.new(prefix: prefix, suffix: suffix, example_prompt: example_prompt, examples: examples, input_variables: input_variables)
       end
 
-      # Load the prompt using the corresponding loader function from the `TYPE_TO_LOADER` hash
-      prompt_loader = TYPE_TO_LOADER[config["_type"]]
-      prompt_loader.call(config)
+      private
+
+      #
+      # Loads the prompt from the given configuration hash
+      #
+      # @param config [Hash] the configuration hash to load from
+      #
+      # @return [Object] the loaded prompt
+      #
+      # @raise [ArgumentError] if the prompt type specified in the config is not supported
+      #
+      def load_from_config(config)
+        # If `_type` key is not present in the configuration hash, add it with a default value of `prompt`
+        unless config.key?("_type")
+          Langchain.logger.warn "No `_type` key found, defaulting to `prompt`"
+          config["_type"] = "prompt"
+        end
+
+        # If the prompt type specified in the configuration hash is not supported, raise an exception
+        unless TYPE_TO_LOADER.key?(config["_type"])
+          raise ArgumentError, "Loading #{config["_type"]} prompt not supported"
+        end
+
+        # Load the prompt using the corresponding loader function from the `TYPE_TO_LOADER` hash
+        prompt_loader = TYPE_TO_LOADER[config["_type"]]
+        prompt_loader.call(config)
+      end
     end
   end
 end

data/lib/langchain/prompt.rb

@@ -0,0 +1,5 @@
+module Langchain
+  module Prompt
+    include Loading
+  end
+end

data/lib/langchain/system_message.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+module Langchain
+  class SystemMessage < Message
+    def type
+      "system"
+    end
+  end
+end

data/lib/langchain/tool/base.rb

@@ -7,16 +7,18 @@ module Langchain::Tool
   #
   # == Available Tools
   #
-  # - {Langchain::Tool::Calculator}: Calculate the result of a math expression
-  # - {Langchain::Tool::RubyCodeInterpretor}: Runs ruby code
+  # - {Langchain::Tool::Calculator}: calculate the result of a math expression
+  # - {Langchain::Tool::Database}: executes SQL queries
   # - {Langchain::Tool::GoogleSearch}: search on Google (via SerpAPI)
+  # - {Langchain::Tool::RubyCodeInterpreter}: runs ruby code
+  # - {Langchain::Tool::Weather}: gets current weather data
   # - {Langchain::Tool::Wikipedia}: search on Wikipedia
   #
   # == Usage
   #
   # 1. Pick the tools you'd like to pass to an Agent and install the gems listed under **Gem Requirements**
   #
-  #     # To use all 3 tools:
+  #     # For example to use the Calculator, GoogleSearch, and Wikipedia:
   #     gem install eqn
   #     gem install google_search_results
   #     gem install wikipedia-client
@@ -28,16 +30,14 @@ module Langchain::Tool
   # 3. Pass the tools when Agent is instantiated.
   #
   #     agent = Langchain::Agent::ReActAgent.new(
-  #       llm: :openai, # or :cohere, :hugging_face, :google_palm or :replicate
-  #       llm_api_key: ENV["OPENAI_API_KEY"],
-  #       tools: ["google_search", "calculator", "wikipedia"]
+  #       llm: Langchain::LLM::OpenAI.new(api_key: "YOUR_API_KEY"), # or other like Cohere, Hugging Face, Google Palm or Replicate
+  #       tools: [
+  #         Langchain::Tool::GoogleSearch.new(api_key: "YOUR_API_KEY"),
+  #         Langchain::Tool::Calculator.new,
+  #         Langchain::Tool::Wikipedia.new
+  #       ]
   #     )
   #
-  # 4. Confirm that the Agent is using the Tools you passed in:
-  #
-  #     agent.tools
-  #     # => ["google_search", "calculator", "wikipedia"]
-  #
   # == Adding Tools
   #
   # 1. Create a new file in lib/langchain/tool/your_tool_name.rb
@@ -53,7 +53,7 @@ module Langchain::Tool
     #
     # @return [String] tool name
     #
-    def tool_name
+    def name
       self.class.const_get(:NAME)
     end
 
@@ -68,7 +68,7 @@ module Langchain::Tool
     #
     # @return [String] tool description
     #
-    def tool_description
+    def description
       self.class.const_get(:DESCRIPTION)
     end
 
@@ -109,7 +109,7 @@ module Langchain::Tool
     #
     def self.validate_tools!(tools:)
       # Check if the tool count is equal to unique tool count
-      if tools.count != tools.map(&:tool_name).uniq.count
+      if tools.count != tools.map(&:name).uniq.count
         raise ArgumentError, "Either tools are not unique or are conflicting with each other"
       end
     end

data/lib/langchain/vectorsearch/milvus.rb

@@ -32,7 +32,7 @@ module Langchain::Vectorsearch
             field: Array(texts)
           }, {
             field_name: "vectors",
-            type: ::Milvus::DATA_TYPES["binary_vector"],
+            type: ::Milvus::DATA_TYPES["float_vector"],
             field: Array(texts).map { |text| llm.embed(text: text) }
           }
         ]
@@ -47,7 +47,7 @@ module Langchain::Vectorsearch
       client.collections.create(
         auto_id: true,
         collection_name: index_name,
-        description: "Default schema created by Vectorsearch",
+        description: "Default schema created by langchain.rb",
         fields: [
           {
             name: "id",
@@ -66,7 +66,7 @@ module Langchain::Vectorsearch
             ]
           }, {
             name: "vectors",
-            data_type: ::Milvus::DATA_TYPES["binary_vector"],
+            data_type: ::Milvus::DATA_TYPES["float_vector"],
             is_primary_key: false,
             type_params: [
               {
@@ -79,6 +79,20 @@ module Langchain::Vectorsearch
       )
     end
 
+    # Create the default index
+    # @return [Boolean] The response from the server
+    def create_default_index
+      client.indices.create(
+        collection_name: "Documents",
+        field_name: "vectors",
+        extra_params: [
+          {key: "metric_type", value: "L2"},
+          {key: "index_type", value: "IVF_FLAT"},
+          {key: "params", value: "{\"nlist\":1024}"}
+        ]
+      )
+    end
+
     # Get the default schema
     # @return [Hash] The response from the server
     def get_default_schema
@@ -91,6 +105,12 @@ module Langchain::Vectorsearch
       client.collections.delete(collection_name: index_name)
     end
 
+    # Load default schema into memory
+    # @return [Boolean] The response from the server
+    def load_default_schema
+      client.collections.load(collection_name: index_name)
+    end
+
     def similarity_search(query:, k: 4)
       embedding = llm.embed(text: query)
 
@@ -101,15 +121,36 @@ module Langchain::Vectorsearch
     end
 
     def similarity_search_by_vector(embedding:, k: 4)
+      load_default_schema
+
       client.search(
         collection_name: index_name,
+        output_fields: ["id", "content", "vectors"],
         top_k: k.to_s,
         vectors: [embedding],
         dsl_type: 1,
         params: "{\"nprobe\": 10}",
-        anns_field: "content",
-        metric_type: "L2"
+        anns_field: "vectors",
+        metric_type: "L2",
+        vector_type: ::Milvus::DATA_TYPES["float_vector"]
       )
     end
+
+    # Ask a question and return the answer
+    # @param question [String] The question to ask
+    # @yield [String] Stream responses back one String at a time
+    # @return [String] The answer to the question
+    def ask(question:, &block)
+      search_results = similarity_search(query: question)
+
+      content_field = search_results.dig("results", "fields_data").select { |field| field.dig("field_name") == "content" }
+      content_data = content_field.first.dig("Field", "Scalars", "Data", "StringData", "data")
+
+      context = content_data.join("\n---\n")
+
+      prompt = generate_prompt(question: question, context: context)
+
+      llm.chat(prompt: prompt, &block)
+    end
   end
 end

data/lib/langchain/vectorsearch/pgvector.rb

@@ -8,7 +8,7 @@ module Langchain::Vectorsearch
     # Gem requirements: gem "pgvector", "~> 0.2"
     #
     # Usage:
-    # pgvector = Langchain::Vectorsearch::Pgvector.new(url:, index_name:, llm:, namespace_column: nil, namespace: nil)
+    # pgvector = Langchain::Vectorsearch::Pgvector.new(url:, index_name:, llm:, namespace: nil)
     #
 
     # The operators supported by the PostgreSQL vector search adapter
@@ -90,20 +90,22 @@ module Langchain::Vectorsearch
     end
 
     # Create default schema
-    # @return [PG::Result] The response from the database
     def create_default_schema
       db.run "CREATE EXTENSION IF NOT EXISTS vector"
-      namespace = namespace_column
+      namespace_column = @namespace_column
       vector_dimension = default_dimension
       db.create_table? table_name.to_sym do
         primary_key :id
         text :content
         column :vectors, "vector(#{vector_dimension})"
-        text namespace.to_sym, default: nil
+        text namespace_column.to_sym, default: nil
       end
     end
 
-    # TODO: Add destroy_default_schema method
+    # Destroy default schema
+    def destroy_default_schema
+      db.drop_table? table_name.to_sym
+    end
 
     # Search for similar texts in the index
     # @param query [String] The text to search for

data/lib/langchain/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Langchain
-  VERSION = "0.6.10"
+  VERSION = "0.6.12"
 end

data/lib/langchain.rb

@@ -3,8 +3,25 @@
 require "logger"
 require "pathname"
 require "colorize"
-
-require_relative "./langchain/version"
+require "zeitwerk"
+loader = Zeitwerk::Loader.for_gem
+loader.ignore("#{__dir__}/langchainrb.rb")
+loader.inflector.inflect(
+  "ai_message" => "AIMessage",
+  "ai21" => "AI21",
+  "ai21_validator" => "AI21Validator",
+  "csv" => "CSV",
+  "html" => "HTML",
+  "json" => "JSON",
+  "jsonl" => "JSONL",
+  "llm" => "LLM",
+  "openai" => "OpenAI",
+  "openai_validator" => "OpenAIValidator",
+  "pdf" => "PDF",
+  "react_agent" => "ReActAgent",
+  "sql_query_agent" => "SQLQueryAgent"
+)
+loader.setup
 
 # Langchain.rb a is library for building LLM-backed Ruby applications. It is an abstraction layer that sits on top of the emerging AI-related tools that makes it easy for developers to consume and string those services together.
 #
@@ -48,13 +65,6 @@ require_relative "./langchain/version"
 #
 # Langchain.logger.level = :info
 module Langchain
-  autoload :Loader, "langchain/loader"
-  autoload :Data, "langchain/data"
-  autoload :Conversation, "langchain/conversation"
-  autoload :ConversationMemory, "langchain/conversation_memory"
-  autoload :DependencyHelper, "langchain/dependency_helper"
-  autoload :ContextualLogger, "langchain/contextual_logger"
-
   class << self
     # @return [ContextualLogger]
     attr_reader :logger
@@ -73,95 +83,7 @@ module Langchain
 
   @root = Pathname.new(__dir__)
 
-  module Agent
-    autoload :Base, "langchain/agent/base"
-    autoload :ReActAgent, "langchain/agent/react_agent/react_agent.rb"
-    autoload :SQLQueryAgent, "langchain/agent/sql_query_agent/sql_query_agent.rb"
-  end
-
-  module Chunker
-    autoload :Base, "langchain/chunker/base"
-    autoload :Text, "langchain/chunker/text"
-    autoload :RecursiveText, "langchain/chunker/recursive_text"
-  end
-
-  module Tool
-    autoload :Base, "langchain/tool/base"
-    autoload :Calculator, "langchain/tool/calculator"
-    autoload :RubyCodeInterpreter, "langchain/tool/ruby_code_interpreter"
-    autoload :GoogleSearch, "langchain/tool/google_search"
-    autoload :Weather, "langchain/tool/weather"
-    autoload :Wikipedia, "langchain/tool/wikipedia"
-    autoload :Database, "langchain/tool/database"
-  end
-
-  module Processors
-    autoload :Base, "langchain/processors/base"
-    autoload :CSV, "langchain/processors/csv"
-    autoload :Docx, "langchain/processors/docx"
-    autoload :HTML, "langchain/processors/html"
-    autoload :JSON, "langchain/processors/json"
-    autoload :JSONL, "langchain/processors/jsonl"
-    autoload :PDF, "langchain/processors/pdf"
-    autoload :Text, "langchain/processors/text"
-    autoload :Xlsx, "langchain/processors/xlsx"
-  end
-
-  module Utils
-    module TokenLength
-      autoload :BaseValidator, "langchain/utils/token_length/base_validator"
-      autoload :AI21Validator, "langchain/utils/token_length/ai21_validator"
-      autoload :CohereValidator, "langchain/utils/token_length/cohere_validator"
-      autoload :GooglePalmValidator, "langchain/utils/token_length/google_palm_validator"
-      autoload :OpenAIValidator, "langchain/utils/token_length/openai_validator"
-      autoload :TokenLimitExceeded, "langchain/utils/token_length/token_limit_exceeded"
-    end
-  end
-
-  module Vectorsearch
-    autoload :Base, "langchain/vectorsearch/base"
-    autoload :Chroma, "langchain/vectorsearch/chroma"
-    autoload :Hnswlib, "langchain/vectorsearch/hnswlib"
-    autoload :Milvus, "langchain/vectorsearch/milvus"
-    autoload :Pinecone, "langchain/vectorsearch/pinecone"
-    autoload :Pgvector, "langchain/vectorsearch/pgvector"
-    autoload :Qdrant, "langchain/vectorsearch/qdrant"
-    autoload :Weaviate, "langchain/vectorsearch/weaviate"
-  end
-
-  module LLM
-    autoload :AI21, "langchain/llm/ai21"
-    autoload :Anthropic, "langchain/llm/anthropic"
-    autoload :Base, "langchain/llm/base"
-    autoload :Cohere, "langchain/llm/cohere"
-    autoload :GooglePalm, "langchain/llm/google_palm"
-    autoload :HuggingFace, "langchain/llm/hugging_face"
-    autoload :LlamaCpp, "langchain/llm/llama_cpp"
-    autoload :OpenAI, "langchain/llm/openai"
-    autoload :Replicate, "langchain/llm/replicate"
-  end
-
-  module Prompt
-    require_relative "langchain/prompt/loading"
-
-    autoload :Base, "langchain/prompt/base"
-    autoload :PromptTemplate, "langchain/prompt/prompt_template"
-    autoload :FewShotPromptTemplate, "langchain/prompt/few_shot_prompt_template"
-  end
-
-  module ActiveRecord
-    autoload :Hooks, "langchain/active_record/hooks"
-  end
-
-  module OutputParsers
-    autoload :Base, "langchain/output_parsers/base"
-    autoload :StructuredOutputParser, "langchain/output_parsers/structured"
-    autoload :OutputFixingParser, "langchain/output_parsers/fix"
-  end
-
   module Errors
     class BaseError < StandardError; end
   end
 end
-
-require "langchain/railtie" if defined?(Rails)